Dockerfile

FROM nvcr.io/nvidia/pytorch:18.03-py3

ARG FFMPEG_VERSION=3.4.2

# nvcuvid deps
RUN apt-get update --fix-missing && \
    apt-get install -y libx11-6 libxext6
ENV NVIDIA_DRIVER_CAPABILITIES=video,compute,utility

# minimal ffmpeg from source
RUN apt-get install -y \
      yasm \
      libx264-148 libx264-dev \
      libx265-79 libx265-dev \
      pkg-config && \
    cd /tmp && wget -q http://ffmpeg.org/releases/ffmpeg-$FFMPEG_VERSION.tar.bz2 && \
    tar xf ffmpeg-$FFMPEG_VERSION.tar.bz2 && \
    rm ffmpeg-$FFMPEG_VERSION.tar.bz2 && \
    cd ffmpeg-$FFMPEG_VERSION && \
    ./configure \
    --prefix=/usr/local \
    --disable-static --enable-shared \
    --disable-all --disable-autodetect --disable-iconv \
    --enable-avformat --enable-avcodec --enable-avfilter --enable-avdevice \
    --enable-protocol=file \
    --enable-demuxer=mov,matroska,image2 \
    --enable-bsf=h264_mp4toannexb,hevc_mp4toannexb \
    --enable-gpl --enable-libx264 --enable-libx265 --enable-zlib \
    --enable-indev=lavfi \
    --enable-swresample --enable-ffmpeg \
    --enable-swscale --enable-filter=scale,testsrc \
    --enable-muxer=mp4,matroska,image2 \
    --enable-cuvid --enable-nvenc --enable-cuda \
    --enable-decoder=h264,h264_cuvid,hevc,hevc_cuvid,png,mjpeg,rawvideo \
    --enable-encoder=h264_nvenc,hevc_nvenc,libx264,libx265,png,mjpeg \
    --enable-hwaccel=h264_cuvid,hevc_cuvid \
    --enable-parser=h264,hevc,png && \
    make -j8 && make install && \
    ldconfig && \
    cd /tmp && rm -rf ffmpeg-$FFMPEG_VERSION && \
    apt-get remove -y yasm libx264-dev libx265-dev && \
    apt-get auto-remove -y

# install stub library since driver libs aren't available at image build time
# this is a temporary requirement that will go away in future cuda versions
# libnvcuvid.so was created using the make-stub.sh script
COPY libnvcuvid.so /usr/local/cuda/lib64/stubs

# install nvvl
RUN pip install --upgrade cmake && \
    apt-get install -y pkg-config && \
    cd /tmp && \
    wget https://github.com/NVIDIA/nvvl/archive/master.tar.gz -O nvvl.tar.gz && \
    mkdir nvvl && \
    tar xf nvvl.tar.gz -C nvvl --strip-components 1 && \
    rm nvvl.tar.gz && \
    cd nvvl/pytorch && \
    python3 setup.py install && \
    pip uninstall -y cmake && \
    apt-get remove -y pkg-config && \
    apt-get autoremove -y

RUN pip install scikit-image tensorflow tensorboard tensorboardX

WORKDIR /workspace/examples/pytorch_superres